Astronomy & Astrophysics manuscript no.

Lya_SFR_SM c
ESO 2018
March 28, 2018

Predicting Lyα escape fractions with a simple observable?

Lyα in emission as an empirically calibrated star formation rate indicator
David Sobral1, 2?? and Jorryt Matthee2

1
Department of Physics, Lancaster University, Lancaster, LA1 4YB, UK
2
Leiden Observatory, Leiden University, P.O. Box 9513, NL-2300 RA Leiden, The Netherlands
March 28, 2018
arXiv:1803.08923v1 [astro-ph.GA] 23 Mar 2018

ABSTRACT

Lyman-α (Lyα) is intrinsically the brightest line emitted from active galaxies. While it originates from many physical processes, for
star-forming galaxies the intrinsic Lyα luminosity is a direct tracer of the Lyman-continuum (LyC) radiation produced by the most
massive O- and early-type B-stars (M? ∼ > 10 M ) with lifetimes of a few Myrs. As such, Lyα luminosity should be an excellent

instantaneous star formation rate (SFR) indicator. However, its resonant nature and susceptibility to dust as a rest-frame UV photon
makes Lyα very hard to interpret due to the uncertain Lyα escape fraction, fesc,Lyα . Here we explore results from the CAlibrating
LYMan-α with Hα (CALYMHA) survey at z = 2.2, follow-up of Lyα emitters (LAEs) at z = 2.2 − 2.6 and a z ∼ 0 − 0.3 compilation
of LAEs to directly measure fesc,Lyα with Hα. We derive a simple empirical relation that robustly retrieves fesc,Lyα as a function of Lyα
rest-frame EW (EW0 ): fesc,Lyα = 0.0048 EW0 [Å] ± 0.05 and we show that the relation is driven by a tight sequence between high
ionisation efficiencies and low dust extinction in LAEs. Observed Lyα luminosities and EW0 are easy measurable quantities at high
redshift, thus making our relation a practical tool to estimate intrinsic Lyα and LyC luminosities under well controlled and simple
assumptions. Our results allow observed Lyα luminosities to be used to compute SFRs for LAEs at z ∼ 0 − 2.6 within ±0.2 dex of
the Hα dust corrected SFRs. We apply our empirical SFR(Lyα,EW0 ) calibration to several sources at z ≥ 2.6 to find that star-forming
LAEs have SFRs typically ranging from 0.1 to 20 M yr−1 and that our calibration might be even applicable for luminous LAEs within
the epoch of re-ionisation. Our results imply higher than canonical ionisation efficiencies and low dust content in LAEs across cosmic
time, and will be easily tested with future observations with JWST which can obtain Hα and Hβ measurements for high-redshift LAEs.
Key words. Galaxies: star formation, starburst, evolution, statistics, general, high-redshift; Ultraviolet: galaxies.

1. Introduction
With a vacuum rest-frame wavelength of 1215.67 Å, the Lyman-
α (Lyα) recombination line (n = 2 → n = 1) plays a key role
in the energy release from ionised hydrogen gas, being intrinsi-
cally the strongest emission line in the rest-frame UV and opti-
cal (e.g. Partridge & Peebles 1967; Pritchet 1994). Lyα is emit-
ted from ionised gas around star-forming regions (e.g. Charlot &
Fall 1993; Pritchet 1994) and AGN (e.g. Miley & De Breuck
2008) and it is routinely used as a way to find high redshift
sources (z ∼ 2 − 7; see e.g. Malhotra & Rhoads 2004).
Several searches for Lyα-emitting sources (Lyα emitters;
LAEs) have led to samples of thousands of star-forming galax-
ies (SFGs) and AGN (e.g. Sobral et al. 2018b, and references
therein). LAEs are typically faint in the rest-frame UV, includ-
ing many that are too faint to be detected by continuum based
searches even with the Hubble Space Telescope (e.g. Bacon et al.
2015). The techniques used to detect LAEs include narrow-band
surveys (e.g. Rhoads et al. 2000; Ouchi et al. 2008; Hu et al.
2010; Matthee et al. 2015), Integral Field Unit (IFU) surveys
(e.g. van Breukelen et al. 2005; Drake et al. 2017) and blind slit
spectroscopy (e.g. Martin & Sawicki 2004; Rauch et al. 2008;
Cassata et al. 2011). Galaxies selected through their Lyα emis-
sion allow for easy spectroscopic follow-up due to their high
?
Based on observations obtained with the Very Large Telescope, pro-
grams: 098.A-0819 & 099.A-0254.
??
e-mail: d.sobral@lancaster.ac.uk
EWs (e.g. Hashimoto et al. 2017) and typically probe low stellar
masses (see e.g. Gawiser et al. 2007; Hagen et al. 2016).
The intrinsic Lyα luminosity is a direct tracer of the ionising
Lyman-continuum (LyC) luminosity and thus a tracer of instan-
taneous star formation rate (SFR), in the same way as Hα is (e.g.
Kennicutt 1998). Unfortunately, inferring intrinsic properties of
galaxies from Lyα observations is extremely challenging. This
is due to the complex resonant nature and sensitivity to dust of
Lyα (see e.g. Dijkstra 2017, for a detailed review on Lyα), which
contrasts with Hα. For example, a significant fraction of Lyα
photons is scattered in the Inter-Stellar Medium (ISM) and in
the Circum-Galactic Medium (CGM) as evidenced by the pres-
ence of extended Lyα halos in LAEs (e.g. Momose et al. 2014;
Wisotzki et al. 2016), but also in the more general population of
z ∼ 2 SFGs sampled by Hα emitters (Matthee et al. 2016), and
the bluer component of such population traced by UV-continuum
selected galaxies (e.g. Steidel et al. 2011). Such scattering leads
to kpc-long random-walks which take millions of years and that
significantly increase the probability of Lyα photons being ab-
sorbed by dust particles. The complex scattering and consequent
higher susceptibility to dust absorption typically leads to low and
uncertain Lyα escape fractions (fesc,Lyα ; the ratio between ob-
served and intrinsic Lyα luminosity; see e.g. Atek et al. 2008).
“Typical" star-forming galaxies at z ∼ 2 have low fesc,Lyα
(∼ 1 − 5%; e.g. Oteo et al. 2015; Cassata et al. 2015), likely
because significant amounts of dust present in their ISM easily
absorb Lyα photons (e.g. Ciardullo et al. 2014; Oteo et al. 2015;
Article number, page 1 of 9
 A&A proofs: manuscript no. Lya_SFR_SM
Oyarzún et al. 2017). However, sources selected through their
Lyα emission typically have ∼ 10 times higher fesc,Lyα (e.g. Song
et al. 2014; Sobral et al. 2017), with Lyα escaping over ≈ 2×
larger radii than Hα (Sobral et al. 2017).
Furthermore, one expects fesc,Lyα to depend on several phys-
ical properties which could be used as predictors of fesc,Lyα . For
example, fesc,Lyα anti-correlates with stellar mass (e.g. Oyarzún
et al. 2017), dust attenuation (e.g. Verhamme et al. 2008; Hayes
et al. 2011; Matthee et al. 2016; An et al. 2017) and SFR (e.g.
Matthee et al. 2016). However, most of these relations require
derived properties (e.g. Yang et al. 2017), show a large scat-
ter, may evolve with redshift and sometimes reveal complicated
trends (e.g. dust dependence; see Matthee et al. 2016).
Interestingly, the Lyα rest-frame equivalent width (EW0 ), a
simple observable, seems to be the simplest direct predictor of
fesc,Lyα in LAEs (Sobral et al. 2017; Verhamme et al. 2017) with
a relation that shows no strong evolution from z ∼ 0 to z ∼ 2 (So-
bral et al. 2017) and that might be applicable at least up to z ∼ 5
(Harikane et al. 2017). Such empirical relation may hold the key
for a simple but useful calibration of Lyα as a direct tracer of the
intrinsic LyC luminosity by providing a way to estimate fesc,Lyα ,
and thus as a good SFR indicator for LAEs (see also Dijkstra
& Westra 2010). We fully explore such possibility and its impli-
cations in this work. In §2 we present the samples at different
redshifts and methods used to compute fesc,Lyα . In §3 we present
and discuss the results, their physical interpretation and our pro-
posed empirical calibration of Lyα as a SFR indicator. Finally,
2.3. Higher redshift LAEs (2.6 ≤ z ≤ 6)
As an application of our results, we explore the publicly avail-
able sample of 3,908 LAEs in the COSMOS field (SC4K sur-
vey; Sobral et al. 2018b) which provides Lyα luminosities and
rest-frame EWs for all LAEs. We also explore published median
or average values for the latest MUSE samples, containing 417
LAEs (e.g. Hashimoto et al. 2017). Note that for all these higher
redshift samples, Hα is not directly available, thus fesc,Lyα cannot
be directly measured (but see Harikane et al. 2017).
2.4. Measuring the Lyα escape fraction ( fesc,Lyα ) with Hα
We use dust corrected Hα luminosity to predict the intrinsic Lyα
luminosity. We then compare the latter to the observed Lyα lu-
minosity to obtain the Lyα escape fraction (fesc,Lyα ). Assuming
case B recombination1 , a temperature of 104 K and an electron
density of 350 cm−3 , we can use the observed Lyα luminosity
(LLyα ), the observed Hα luminosity (LHα ) and the dust extinc-
tion affecting LHα (AHα 2 , in mag) to compute fesc,Lyα as:
LLyα
fesc,Lyα = . (1)
8.7 LHα × 100.4×AHα
This means that with our assumptions so far, and provided that
we know fesc,Lyα , we can use the observed LLyα to obtain the
intrinsic Hα luminosity. Therefore, one can use Lyα as a star
formation rate (SFR) indicator3 following Kennicutt (1998) for
we present the conclusions in §4. We adopt a flat cosmology with a Salpeter (Chabrier) IMF (0.1 − 100 M ):
Ωm = 0.3, ΩΛ = 0.7, and H0 = 70 km s−1 Mpc−1 .
7.9(4.4) × 10−42 LLyα
2. Sample and Methods
2.1. LAEs at low redshift (z ≤ 0.3)
For our lower redshift sample, we explore a compilation of 30
sources presented in Verhamme et al. (2017) which have ac-
curate (Hα derived) fesc,Lyα measurements and sample a range
of galaxy properties. The sample includes high EW Hα emit-
ters (HAEs) from the Lyman Alpha Reference Sample at z =
0.02 − 0.2 (LARS, e.g. Hayes et al. 2013, 2014), a sample of
LyC leakers (LyCLs) investigated in Verhamme et al. (2017) at
z ∼ 0.3 (Izotov et al. 2016a,b) and a more general ‘green pea’
(GPs) sample (e.g. Cardamone et al. 2009; Henry et al. 2015;
Yang et al. 2016, 2017). These are all LAEs at low redshift with
available Lyα, Hα and dust extinction information required to es-
timate fesc,Lyα (see §2.4) and for which Lyα EW0 s are available.
For more details on the sample, see Verhamme et al. (2017) and
references therein.
2.2. LAEs at cosmic noon (z = 2.2 − 2.6)
For our sample at the peak of star formation history we use 188
narrow-band selected LAEs with Hα measurements from the
CALYMHA survey at z = 2.2 (Matthee et al. 2016; Sobral et al.
2017) presented in Sobral et al. (2017), for which fesc,Lyα mea-
surements are provided as a function of EW0 . In addition, we
explore spectroscopic follow-up of CALYMHA sources with X-
SHOOTER on the VLT (Sobral et al. 2018a) and individual mea-
surements for four sources (CALYMHA-67, -93, -147 and -373;
see Sobral et al. 2018a). For those sources we measure Lyα, Hα
and Hβ. Furthermore, we also use a sample of 29 narrow-band
selected LAEs at z ∼ 2.6 presented by Trainor et al. (2016), for
which Lyα and Hα measurements are available.
Article number, page 2 of 9
SFRLyα [M yr−1 ] = (2)
(1 − fesc,LyC ) 8.7 fesc,Lyα
where fesc,LyC is the escape fraction of ionising LyC photons (see
e.g. Sobral et al. 2018b). In practice, fesc,LyC is typically assumed
to be ≈ 0, but it may be ≈ 0.1 − 0.15 for LAEs (see discussions
in e.g. Matthee et al. 2017a; Verhamme et al. 2017).
2.5. Statistical fits and errors
For all fits and relations in this work (e.g. fesc,Lyα vs. EW0 ), we
vary each data-point within its full Gaussian probability distribu-
tion function independently (both in EW0 and fesc,Lyα ), and re-fit
10,000 times. We present the best-fit relation as the median of
all fits, and the uncertainties (lower and upper) are the 16 and
84 percentiles. For bootstrapped quantities (e.g. for fitting the
low redshift sample) we obtain 10,000 samples randomly pick-
ing half of the total number of sources and computing that spe-
cific quantity. We fit relations in the form y = Ax + B.
3. Results and Discussion
3.1. The observed fesc,Lyα -EW0 relation at z ∼ 0.1 − 2.6
Figure 1 shows that fesc,Lyα correlates with Lyα EW0 with ap-
parently no redshift evolution between z = 0 − 2.6 (see also
Verhamme et al. 2017; Sobral et al. 2017). We find that fesc,Lyα
varies continuously from ≈ 0.2 to ≈ 0.7 for LAEs from the low-
est (≈ 30 Å) to the highest (≈ 120 − 160 Å) Lyα rest-frame EWs.
1
We use Lyα/Hα = 8.7, but vary the Lyα/Hα case B ratio between 8.0
and 9.0 to test for its effect; see §3.5.
2
With our case B assumptions the intrinsic Balmer decrement is:
Hα/Hβ = 2.86. Using a Calzetti et al. (2000) dust law we use AHα =
6.531 log10 (Hα/Hβ) − 2.981 (see details in e.g. Sobral et al. 2012).
3
For continuous star formation over 10 Myr timescales.
 Sobral & Matthee: Predicting Lyα escape fractions with a simple observable

1.0 z = 2.2 LAEs (S+17)

z = 2.6 LAEs (T+16)

Ly↵ escape fraction (fesc,Ly↵)

1
0.8

=
⇠ion

UV
V)

2
0.6

=
2. 6

E(B

UV
2. 2 0. 3
z⇠ 0. 1

0)
z⇠

+1
0.4
W
(D
UV
Typical uncertainty
m
fro
ed

0.2
ict
ed

z ⇠ 0.1 HAEs (H+13)

Pr

z ⇠ 0.3 LyCLs (V+17)

z ⇠ 0.3 GPs (Y+16)
0.0
0 20 40 60 80 100 120 140 160
Ly↵ EW0 (Å, rest-frame)
Fig. 1. The relation between fesc,Lyα and Lyα EW0 for z ∼ 2.2 (stacks; see Sobral et al. 2017), z ∼ 2.6 (binning; Trainor et al. 2015) and comparison
with z ∼ 0 − 0.3 samples (e.g. Cardamone et al. 2009; Hayes et al. 2013; Henry et al. 2015; Yang et al. 2016, 2017; Verhamme et al. 2017),
estimated from dust-corrected Hα luminosities (Equation 1). We show the 1 σ and 2 σ range for the fits at z ∼ 2.2 − 2.6 and z ∼ 0 − 0.3 separately,
and find them to be consistent within those uncertainties, albeit with a potential steeper relation at higher redshift. We find a combined best fitting
relation given by fesc,Lyα = 0.0048 EW0 ± 0.05. The observed relation is significantly away from what would be predicted (DW+10) based on the
UV (see Dijkstra & Westra 2010), and implies not only a higher ξion than the canonical value, but also an increasing ξion as a function of EW0 .

We use our samples at z ∼ 0 − 0.3 and z ∼ 2.2 − 2.6, separately Table 1. The results from fitting the relation between fesc,Lyα and Lyα
and together, to obtain linear fits to the relation between fesc,Lyα EW0 as fesc,Lyα = A × EW0 + B, with EW0 in Å (see §2.5). [i: individual
and Lyα EW0 (see §2.5). These fits allow us to provide a more sources used for fitting; b: binned/averaged quantity used for fitting; B:
quantitative view on the empirical relation and evaluate any sub- bootstrap analysis when fitting each of the 10,000 times; G: each data
bin is perturbed along its Gaussian probability distribution.]
tle redshift evolution; see Table 1.
The relation between fesc,Lyα and Lyα EW0 is statistically sig- Sample A (Å−1 ) B [notes]
nificant at 5 to 10 σ for all redshifts. We note that all linear fits z ∼ 0 − 0.3 0.0041+0.0006
−0.0004 0.00+0.03
−0.02 [i,B]
are consistent with a zero escape fraction for a null EW0 (Table z ∼ 2.2 0.0056+0.0012
−0.0011 0.00+0.05
−0.05 [b,G]
1), suggesting that the trend is well extrapolated for weak LAEs z ∼ 2.6 0.0054+0.0016
−0.0015 0.01+0.11
−0.11 [b,G]
with EW0 ≈ 0 − 20 Å. Furthermore, as Table 1 shows, the fits to z ∼ 0 − 2.2 0.0045+0.0008
−0.0007 0.00+0.06
−0.06 [b,G]
the individual (perturbed) samples at different redshifts result in z ∼ 2.2 − 2.6 0.0056+0.0012 0.00+0.07 [b,G]
relatively similar slopes and normalisations within the uncertain- −0.0012 −0.08
z ∼ 0 − 2.6 0.0048+0.0007 0.00+0.05 [b,G]
ties, and thus are consistent with the same relation from z ∼ 0 to −0.0007 −0.05
z ∼ 2.6. Nevertheless, we note that there is minor evidence for a
shallower relation at lower redshift for the highest EW0 (Figure
1), but this could be driven by current samples selecting sources efficiency of ionising photons compared to the UV luminosity,
with more extreme properties (including LyC leakers). Given our ξion (see e.g. Matthee et al. 2017a; Shivaei et al. 2017).
findings, we decide to combine the samples and obtain joint fits,
with the results shown in Table 1. The slope of the relation is While a relation between fesc,Lyα and EW0 is expected, we
consistent with being ≈ 0.005 with a null fesc,Lyα for EW0 = 0 Å. can investigate if it simply follows what would be predicted
given that both the UV and Lyα trace SFRs. In order to predict
fesc,Lyα based on Lyα EW0 we follow Dijkstra & Westra (2010)
3.2. The fesc,Lyα -EW0 relation: expectation vs. reality who use the Kennicutt (1998) SFR calibrations for a Salpeter
IMF. As in Dijkstra & Westra (2010), we assume two different
The existence of a relation between observed Lyα luminosity UV slopes: β = −2.0 and β = −1.0, which encompass the ma-
and EW0 (Figure 1) is not surprising. This is because Lyα EW0 jority of LAEs (note that a steeper β results in an even more
is sensitive to the ratio between Lyα and the UV, which can be significant disagreement with observations) and can predict that
νLyα −2−β
used as a proxy of the fesc,Lyα (see Sobral et al. 2018b). However, fesc,Lyα = C × EW E , with E = 76 Å and C = νUV
0
. We use
the slope, normalisation and scatter of such relation depend on C = 0.89 and C = 0.75 for the different β slopes as in Dijkstra
complex physical conditions such as dust obscuration, differen- & Westra (2010). Note that this methodology implicitly results
tial dust geometry, scattering of Lyα photons and the production in assuming a “canonical", constant ξion = 1.3 × 1025 Hz erg−1
Article number, page 3 of 9

1.0
Ly↵ escape fraction (fesc,Ly↵)
V) 5
1
0.8 0.0
=
UV
E(B
0.6
=
UV
1
0.
0.4
0.2
0.3
0.0
0 20 40 60 80 100 120 140 160
Ly↵ EW0 (Å, rest-frame)
Fig. 2. Left: The predicted relation between fesc,Lyα and Lyα EW0 for different E(B− V) (contour levels) with our toy model (see §3.3 and Appendix
A). We find that dust extinction drives the simple predicted relation down, with data at z ∼ 0 − 2.6 hinting for lower dust extinction at the highest
EW0 and higher dust extinction at the lowest EW0 , but with the range being relatively small overall and around E(B − V) ≈ 0.1 − 0.2. Right: The
predicted relation between fesc,Lyα and Lyα EW0 by varying ξion (contours). We find that while increasing E(B − V) mostly shifts the relation down,
increasing ξion moves the relation to the right. Observations thus hint for an increase in the typical ξion for LAEs with increasing EW0 .
(Kennicutt 1998)4 , and a unit ratio between Lyα and UV SFRs
(see Sobral et al. 2018b).
Predicting fesc,Lyα based on the ratio of Lyα to UV using EW0
(see Dijkstra & Westra 2010) significantly overestimates fesc,Lyα
(as indicated by the dot-dashed lines in Figure 1). Observations
reveal higher Lyα EW0 (by a factor of just over ∼ 2) than ex-
pected for a given fesc,Lyα , with the offset between the simple pre-
diction and observations potentially becoming larger for increas-
ing EW0 . These results reveal processes that can boost the ratio
between Lyα and UV (boosting EW0 ), particularly by boosting
Lyα, or processes that reduce fesc,Lyα . Potential explanations in-
clude scattering, (differential) dust extinction, excitation due to
shocks originating from stellar winds and/or AGN activity, and
short time-scale variations in SFRs, leading to a higher ξion (see
Figure 1). High ξion values (ξion ≈ 3 × 1025 Hz erg−1 ) seem to
be typical for LAEs (e.g. Matthee et al. 2017a; Nakajima et al.
2018) and may explain the observed relation, even more so if ξion
rises with increasing EW0 (e.g. Matthee et al. 2017a), but dust
extinction likely also plays a role (Figure 1).
3.3. The fesc,Lyα -EW0 relation: physical interpretation
In order to further interpret the role of dust (E(B−V)) and ξion on
the observed fesc,Lyα -EW0 and what the relation may be telling
us, we produce a simple analytical model (see details in Ap-
pendix A). We independently vary SFRs, E(B − V) and ξion . The
toy model follows our framework using a Calzetti et al. (2000)
dust law and the Kennicutt (1998) calibrations and relations be-
tween UV and Hα. We also vary some assumptions indepen-
dently, which include the intrinsic Lyα/Hα ratio and fesc,LyC . Fur-
thermore, we introduce an extra parameter to further vary fesc,Lyα
and mimic processes which are hard to model, such as scattering,
which can significantly reduce or even boost fesc,Lyα (Neufeld
1991). We compute observed Lyα EW0 and compare them with
fesc,Lyα for 20,000 galaxy realisations. Further details are given
in Appendix A.
The key results from our toy model are shown in Figure 2.
We find that both E(B − V) and ξion likely play a role in setting
4
ξion = 1.3 × 1025 SFR
SFRHα
UV
(Hz erg−1 ).
Article number, page 4 of 9
A&A proofs: manuscript no. Lya_SFR_SM
1.0
24.8
25.0
Ly↵ escape fraction (fesc,Ly↵)
0.8
2
25.
log10 ⇠ion
2
1
0.6
=
=
UV
.4
UV
25
0.4
0.15
z = 2.6 LAEs (T+16) z = 2.6 LAEs (T+16)
0.2 z = 2.2 LAEs (S+17) 0.2 z = 2.2 LAEs (S+17)
z ⇠ 0.3 LyCLs (V+17) z ⇠ 0.3 LyCLs (V+17)
z ⇠ 0.3 GPs (Y+16) z ⇠ 0.3 GPs (Y+16)
z ⇠ 0.1 HAEs (H+13) z ⇠ 0.1 HAEs (H+13)
0.0
0 20 40 60 80 100 120 140 160
Ly↵ EW0 (Å, rest-frame)
the fesc,Lyα -EW0 relation and changing it from simple predictions
to the observed relation (see §3.2). As the left panel of Figure
2 shows, observed LAEs on the fesc,Lyα -EW0 relation seem to
have low E(B − V) ≈ 0.1 − 0.2, with the lowest EW0 sources
displaying typically higher E(B − V) of 0.2-0.3 and the highest
EW0 sources likely having lower E(B − V) of < 0.1. Further-
more, as the right panel of Figure 2 shows, high EW0 LAEs have
higher ξion , potentially varying from log10 (ξion /Hz erg−1 ) ≈ 25 to
log10 (ξion /Hz erg−1 ) ≈ 25.4 − 25.5. Our toy model interpretation
is consistent with recent results (e.g. Trainor et al. 2016; Matthee
et al. 2017a; Nakajima et al. 2018) for high EW0 LAEs. Overall,
a simple way to explain the fesc,Lyα -EW0 relation at z ∼ 0 − 2.6
is for LAEs to have narrow ranges of low E(B − V) ≈ 0.1 − 0.2,
that decrease slightly as a function of EW0 and a relatively nar-
row range of high ξion values that increase with EW0 .
Our toy model explores the full range of physical conditions
independently without making any assumptions on how param-
eters may correlate, in order to interpret the observations in a
simple unbiased way. However, the fact that observed LAEs fol-
low a relatively tight relation between fesc,Lyα and EW0 suggests
that there are important correlations between e.g. dust, age and
ξion . By selecting simulated sources in our toy model that lie on
the observed relation (see Appendix A.1), we recover a tight cor-
relation between ξion and E(B − V), while the full generated pop-
ulation in our toy model shows no correlation at all by definition
(see Figure A.1). This implies that the observed fesc,Lyα -EW0 is
likely a consequence of an evolutionary ξion -E(B − V) sequence
for LAEs. For further details, see Appendix A.1.
3.4. Estimating fesc,Lyα with a simple observable: Lyα EW0
We find that LAEs follow a simple relation between fesc,Lyα and
Lyα EW0 roughly independently of redshift (for z ≤ 2.6). Moti-
vated by this, we propose the following empirical estimator (see
Table 1) for fesc,Lyα as a function of Lyα EW0 (Å):
fesc,Lyα = 0.0048+0.0007
−0.0007 EW0 ± 0.05 [ 0 < EW0 < 160 ]. (3)
This relation may hold up to EW0 ≈ 210 Å, above which we
would predict fesc,Lyα ≈ 1. This relation suggests that it is pos-
sible to estimate fesc,Lyα for LAEs within 0.2 dex even if only
 Sobral & Matthee: Predicting Lyα escape fractions with a simple observable

0.8
z = 2.2 LAEs (S+17) LLy↵ ⇥7.9⇥10 42

log10(SFRLy↵(This work)/SFRH↵)
SFRLy↵ =
0.6 z = 2.6 LAEs (T+16) (1 fesc,LyC )(0.042 EW0 )
z = 4.9 LAEs (H+17)
0.4 Typical SFR scatter z ⇠ 0 2.2 (UV-H↵ or FIR-H↵)

0.2
p
0.0 / N

(±30 Å)
-0.2
-0.4
-0.6 z ⇠ 0.1 HAEs (H+13) z ⇠ 0.3 GPs (Y+16)
z ⇠ 0.3 LyCLs (V+17) z = 2.2 LAEs (S+18)
-0.8
0 20 40 60 80 100 120 140 160
Ly↵ EW0 (Å, rest-frame)
Fig. 3. The logarithmic ratio between SFRs computed with Equation 4 using Lyα luminosity and EW0 and the “true" SFR, measured directly
from dust-corrected Hα luminosity (given our definitions, log10 (SFRLyα(This work) /SFRHα ) = log10 (fesc,Lyα(Hα) /fesc,Lyα(This work) )). We find a relatively
small scatter which may decrease for higher EWs and that is at the global level of ±0.12 dex for the typical definition of LAE at higher redshift
(EW0 > 20 Å), but rises to ≈ 0.2 dex at the lowest EWs, likely due to a larger range of dust properties. We also provide a comparison of the typical
scatter between UV and FIR SFRs in relation to Hα at z ∼ 0 − 2 (≈ 0.3 dex; see e.g. Domínguez Sánchez et al. 2012; Oteo et al. 2015).

the Lyα EW0 is known/constrained. It also implies that the ob- 3.5. Lyα as a SFR indicator: empirical calibration and errors
served Lyα luminosities are essentially equal to intrinsic Lyα
Driven by the simple relation found up to z ∼ 2.6, we derive an
luminosities for sources with EW0 as high as ≈ 200 Å. We pro-
empirical calibration to obtain SFRs based on two simple, direct
pose a linear relation for its simplicity and because current data
observables for LAEs at high redshift: 1) Lyα EW0 and 2) ob-
do not suggest a more complex relation. Larger data-sets with
served Lyα luminosity. This calibration is based on observables,
Hα and Lyα measurements, particularly those covering a wider
but predicts the dust-corrected SFR. Based on Equations 2 and
parameter space may lead to the necessity of a more complicated
3, for a Salpeter (Chabrier) IMF we can derive6 :
functional form. A departure from a linear fit may also provide
further insight of different physical processes driving the relation LLyα × 7.9 (4.4) × 10−42
(e.g. winds, orientation angle, burstiness or additional ionisation
processes such as fluorescence).
We further test the validity of Equation 3 by measuring the
ratio between the real (Hα-based) fesc,Lyα fraction and that in-
ferred from the simple predicting relation. We conclude that
while the escape of Lyα photons can depend on a range of prop-
erties in a very complex way (see e.g. Hayes et al. 2010; Matthee
et al. 2016; Yang et al. 2017), using EW0 and Equation 3 leads
to predicting fesc,Lyα within ≈ 0.1 − 0.2 dex of real values. This
compares with a larger scatter of ≈ 0.3 dex for relations with
derivative or more difficult quantities to measure such as dust ex-
tinction or the red peak velocity of the Lyα line (e.g. Yang et al.
2017). Equation 3 may thus be applied to estimate fesc,Lyα for a
range of LAEs in the low and higher redshift Universe. For ex-
ample, J1154+2443 (Izotov et al. 2018), has a measured fesc,Lyα
directly from dust corrected Hα luminosity of ≈ 0.7−0.85 , while
Equation 3 would imply ≈ 0.6 − 0.7 based on the EW0 ≈ 133 Å
for Lyα, thus implying a difference of only 0.06-0.1 dex. Further-
more, in principle, Equation 3 could also be explored to trans-
form EW0 distributions (e.g. Hashimoto et al. 2017, and refer-
ences therein) into distributions of fesc,Lyα for LAEs.
5
This may be up to ≈ 0.98 if Hβ is used; see (Izotov et al. 2018).
SFRLyα [M yr−1 ] = (±15%) (4)
(1 − fesc,LyC )(0.042 EW0 )
The current best estimate of the scatter in Equation 3 (the
uncertainty in the relation to calculate fesc,Lyα is ±0.05) implies a
±0.07 dex uncertainty in the extinction corrected SFRs from Lyα
with our empirical calculation. In order to investigate other sys-
tematic errors, we conduct a Monte Carlo analysis by randomly
varying fesc,LyC (0.0 to 0.2) and the case B coefficient (from 8.0
to 9.0), along with perturbing fesc,Lyα from −0.05 to +0.05. We
assume that all properties are independent, and thus this can be
seen as a conservative approach to estimate the uncertainties. We
find that the uncertainty in fesc,Lyα is the dominant source of un-
certainty (12%) with the uncertainty on fesc,LyC and the case B
coefficient contributing an additional 3% for a total of 15%. This
leads to an expected uncertainty of Equation 4 of 0.08 dex.
3.6. Lyα as a SFR indicator: performance and implications
In Figure 3 we apply Equation 4 to compare the estimated SFRs
(from Lyα) with those computed with dust corrected Hα lumi-
nosities. We also include individual sources at z ∼ 2.2 (S18; So-
bral et al. 2018a) and recent results from Harikane et al. (2017) at
6
Note that the constant 0.042 has units of Å−1 , and results from 8.7 ×
0.0048 Å−1 .
Article number, page 5 of 9
 A&A proofs: manuscript no. Lya_SFR_SM

z = 4.8 which were not used in the calibration, and thus provide

log10 (SFRLy↵ /SFRUV )

0.6
an independent way to test our new calibration. We find a global Typical SFR scatter at z ⇠ 2
0.3
scatter of ≈ 0.12 dex, being apparently larger for lower EW0 ,
but still lower than the typical scatter between SFR indicators 0.0
after dust corrections (e.g. UV-Hα or FIR-Hα; see Domínguez -0.3
Sánchez et al. 2012; Oteo et al. 2015), as shown in Figure 3. -0.6 AUV = 1 mag
The small scatter and approximately null offset between our cal- 10 20 40 80 160

ibration’s prediction and measurements presented by Harikane

160
ion
et al. (2017) at z ∼ 5 suggest that Equation 4 may be applicable

SFRUV+AUV (M yr 1)
at
ul
at higher redshift with similarly competitive uncertainties (see
§3.7 and §3.8). 80 im
S

AUV = 1 mag
3.7. Application to bright and faint LAEs at high redshift
40
Our new empirical calibration of Lyα as a SFR indicator allows
to estimate SFRs of LAEs at high redshift. The global Lyα lu-
20
minosity function at z ∼ 3 − 6 has a typical Lyα luminosity ion
of 1042.9 erg s−1 (Sobral et al. 2018b), with these LAEs having lat
re
1
EW0 ≈ 80 Å (suggesting fesc,Lyα = 0.38 ± 0.05 with Equation 3), 1:
10 z⇠6 8 UV-selected (M+17)
which implies SFRs of ≈ 20 M yr−1 . If we explore the public z⇠6 7 Ly↵-selected (M+17)
SC4K sample of LAEs at z ∼ 2−6 (Sobral et al. 2018b), limiting
it to sources with up to EW0 = 210 Å and that are consistent with 10 20 40 80 160
being star-forming galaxies (LLyα < 1043.2 erg s−1 ; see Sobral SFRLy↵, EW0 (M yr 1) this study
et al. 2018a), we find a median SFR for LAEs of 12+9 −1
−5 M yr ,
−1 −1
ranging from ≈ 2 M yr to ≈ 90 M yr at z ∼ 2 − 6. These Fig. 4. Comparison between SFRs computed with our new empirical
calibration for Lyα as a SFR indicator (Equation 4) and those computed
reveal that “typical" to luminous LAEs are forming stars below based on dust corrected UV luminosity (see §3.8) for a compilation of
and up to the typical SFR (SFR? ≈ 40 − 100 M yr−1 ) at high
redshift (see Smit et al. 2012; Sobral et al. 2014).
Deep MUSE Lyα surveys (e.g. Drake et al. 2017; Hashimoto
et al. 2017) are able to sample the faintest LAEs with a median
LLyα = 1041.9±0.1 erg s−1 and EW0 = 87 ± 6 (Hashimoto et al.
2017) at z ∼ 3.6. We predict a typical fesc,Lyα = 0.42 ± 0.05 and
z ∼ 5 − 8 sources (see Matthee et al. 2017c, and references therein).
Our simple empirical calibration of Lyα as a SFR is able to recover dust
corrected UV SFRs with a typical scatter of ≈ 0.2 dex, being slightly
higher for the more luminous LAEs than for the continuum selected
LAEs which probe down to lower SFRs (scatter ≈ 0.08 dex which is
very close to the systematic scatter expected; see §3.5). We also com-
SFRLyα = 1.7±0.3 M yr−1 for those MUSE LAEs. Furthermore, pute SFRs in the same way with observables from our toy model and
show the results of our simulation. We find that the scatter in our toy
the faintest LAEs found with MUSE have LLyα = 1041 erg s−1 model is much larger, with this being driven by E(B − V) being able to
(Hashimoto et al. 2017), implying SFRs of ≈ 0.1 M yr−1 with
our calibration. Follow-up JWST observations targeting the Hα
line for faint MUSE LAEs are thus expected to find typical Hα
luminosities of 2×1041 erg s−1 and as low as ≈ 1−2×1040 erg s−1
for the faintest LAEs. Based on our predicted SFRs, we expect
MUSE LAEs to have UV luminosities from MUV ≈ −15.5 for
the faintest sources, to MUV ≈ −19 for more typical LAEs, thus
potentially linking faint LAEs discovered from the ground with
the population of SFGs that dominate the faint end of the UV
luminosity function (e.g. Fynbo et al. 2003; Gronke et al. 2015;
Dressler et al. 2015).
3.8. Comparison with UV and implications at higher redshift
Equations 3 and 4 can be applied to a range of spectroscopically
confirmed LAEs in the literature. We also extend our predictions
to sources within the epoch of re-ionisation7 . We explore a recent
extensive compilation by Matthee et al. (2017c) of both Lyα-
and UV-selected LAEs with spectroscopic confirmation and Lyα
measurements (e.g. Ouchi et al. 2008, 2009; Ono et al. 2012; So-
bral et al. 2015; Zabl et al. 2015; Stark et al. 2015c; Ding et al.
2017; Shibuya et al. 2018). These include published LLyα , EW0
and MUV . In order to correct UV luminosities we use the UV β
slope, typically used to estimate AUV 8 . For UV-selected sources
we assume β = −1.6 ± 0.2 dex (typical for their UV luminosity;
7
See Laursen et al. (2011) for important caveats on how the transmis-
sion at line-centre is affected by an increasing IGM neutral fraction
8
We use AUV = 4.43 + 1.99β; see Meurer et al. (1999).
Article number, page 6 of 9
vary from 0.0 to 0.5.
e.g. Bouwens et al. 2009), while for the luminous LAEs we use
β = −1.9 ± 0.2 dex. We predict their SFRs using LLyα and EW0
only (Equation 4) and compare with SFRs measured from dust-
corrected UV luminosities (Kennicutt 1998); see Table A.2. We
make the same assumptions and follow the same methodology
to transform the observables of our toy model into SFRs (see
Figure 4). We note that, as our simulation shows, one expects a
correlation even if our calibration of Lyα as a SFR indicator is
invalid at high redshift. Therefore, we focus our discussion on
the normalisation of the relation and particularly on the scatter,
not on the existence of a relation. We also note that our calibra-
tion is based on dust corrected Hα luminosities at z ∼ 0 − 2.6,
and that UV luminosities are not used prior to this Section.
Our results are shown in Figure 4 (see Table A.2 for details
on individual sources), which contains sources at a variety of
redshifts, from z ∼ 6 to z ∼ 8 (e.g. Oesch et al. 2015; Stark et al.
2017). We find a remarkable agreement between our predicted
Lyα SFRs based solely on Lyα luminosities and EW0 and the
dust corrected UV SFRs for a range of sources at z ∼ 6 − 8. We
find that the scatter between UV-based and Lyα based SFRs to
be ≈ 0.2 dex. Interestingly, we find a larger scatter for sources
selected as LAEs (0.23 dex) than those that were selected using
UV continuum using e.g. HST (although they are also LAEs),
for which we find a scatter of only 0.08 dex.
Overall, our results and application to higher redshift reveals
that Equation 4 is able to retrieve SFRs with very simple observ-
 Sobral & Matthee: Predicting Lyα escape fractions with a simple observable
ables even for LAEs within re-ionisation (e.g. Ono et al. 2012;
Stark et al. 2015c, 2017; Schmidt et al. 2017). In the early Uni-
verse the fraction of sources that are LAEs is higher, thus mak-
ing our calibration applicable to a larger fraction of the galaxy
population, perhaps with an even smaller scatter due to the ex-
pected narrower range of physical properties. Our calibration of
Lyα as a SFR indicator is simple, directly calibrated with Hα,
and should not have a significant dependence on e.g. metallicity,
unlike other proposed SFRs tracers at high redshift such as [Cii]
luminosity or other weak UV metal lines.
It is nonetheless surprising that our calibration apparently
still works even at z ∼ 7−8 for luminous LAEs. This seems to in-
dicate that the IGM may not play a significant role for these Lyα-
visible sources, potentially due to early ionised bubbles (Matthee
et al. 2015) or velocity offsets of Lyα with respect to systemic
(see e.g. Stark et al. 2017).
3.9. A tool for re-ionisation: predicting the LyC luminosity
Based on our results and assumptions (see §2.4), we follow
Matthee et al. (2017a)9 and derive a simple expression to pre-
dict the number of produced LyC photons per second, Qion (s−1 )
with direct Lyα observables (LLyα and EW0 ):
LLyα
This allows the prediction of fesc,Lyα based on a simple direct
observable, and thus to compute the intrinsic Lyα luminosity
of LAEs at high redshift.
• The observed fesc,Lyα -EW0 implies a tight ξion -E(B − V) se-
quence for LAEs, with higher ξion at lower E(B − V) and
vice versa. Both ξion and E(B − V) seem to depend on Lyα
EW0 (Figure 2). Our results imply that the higher the EW0
selection, the higher the ξion and the lower the E(B − V).
• The fesc,Lyα -EW0 relation reveals a scatter of only 0.1-0.2 dex
for LAEs, and there is evidence for the relation to hold up to
z ∼ 5 (Figure 3). The scatter is higher towards lower EW0 ,
consistent with a larger range in dust properties for sources
with the lowest EW0 . At the highest EW0 , on the contrary,
the scatter may be as small as ≈ 0.1 dex, consistent with high
EW0 LAEs being an even more homogeneous population of
dust-poor, high ionisation star-forming galaxies.
• We use our results to calibrate Lyα as a SFR indicator for
LAEs (Equation 4) and find a global scatter of 0.2 dex be-
tween measurements using Lyα only and those using dust-
corrected Hα luminosities. Our results also allow us to derive
a simple estimator of the number of LyC photons produced
per second (Equation 5) with applications to studies of the
epoch of re-ionisation.
• Equation 4 implies that star-forming LAEs at z ∼ 2 − 6 have
Qion,Lyα [s−1 ] = , (5) SFRs typically ranging from 0.1 to 20 M yr−1 , with MUSE
cHα (1 − fesc,LyC ) (0.042 EW0 )
LAEs expected to have typical SFRs of 1.7 ± 0.3 M yr−1 ,
where cHα = 1.36 × 10−12 erg (e.g. Kennicutt 1998; Schaerer and more luminous LAEs having SFRs of 12+9 −1
2003), under our case B recombination assumption (see §2.4).
Recent work by e.g. Verhamme et al. (2017) show that LyC
leakers are strong LAEs, and that fesc,Lyα is linked and/or can
be used to predict fesc,LyC (see Chisholm et al. 2018). Equation
5 provides an extra useful tool: an empirical simple estimator
of Qion for LAEs given observed Lyα luminosities and EW0 .
Note that Equation 5 does not require measuring UV luminosi-
ties or ξion , but instead direct, simple observables. Matthee et al.
(2017c) already used a similar method to predict ξion at high red-
shift. Coupled with an accurate estimate of the escape fraction
of LyC photons from LAEs, which can be obtained with HST,
a robust estimate of the full number density of LAEs from faint
to the brightest sources (Sobral et al. 2018b) and their redshift
evolution, Equation 5 may provide a simple tool to further un-
derstand if LAEs were able to re-ionise the Universe.
4. Conclusions
Lyα is intrinsically the brightest emission-line in active galaxies,
and should be a good SFR indicator. However, the uncertain and
difficult to measure fesc,Lyα has limited the interpretation and use
of Lyα luminosities. In order to make progress, we have explored
samples of LAEs at z = 0 − 2.6 with direct Lyα escape fractions
measured from dust corrected Hα luminosities which do not re-
quire any SED fitting, ξion or other complex assumptions based
on derivative quantities. Our main results are:
• There is a simple, linear relation between fesc,Lyα and Lyα
EW0 : fesc,Lyα = 0.0048 EW0 [Å] ± 0.05 (Equation 3) which is
shallower than simple expectations, due to both more ionis-
ing photons per UV luminosity (ξion ) and declining dust ex-
tinction (E(B−V)) for LAEs with increasing EW0 (Figure 1).
9 Bouwens, R. J., Illingworth, G. D., Franx, M., et al. 2009, ApJ, 705, 936
We assume fdust ≈ 0 (see Matthee et al. 2017a), i.e., we make the
assumption that for LAEs the dust extinction to LyC photons within
HII regions is ≈ 0.
−5 M yr .
• SFRs based on Equation 4 are in very good agreement
with dust corrected UV SFRs even within the epoch of re-
ionisation and for a range of sources, hinting for it to be ap-
plicable in the very early Universe. If shown to be the case,
our results have implications for the minor role of the IGM in
significantly changing Lyα luminosities and EW0 for lumi-
nous LAEs within the epoch of re-ionisation, and show that
measuring LLyα and EW0 provide apparently reliable SFRs.
Our results provide a simple interpretation of the tight fesc,Lyα -
EW0 relation. Most importantly, we provide simple and practical
tools to estimate fesc,Lyα at high redshift with two direct observ-
ables and thus to use Lyα as a SFR indicator and to measure the
number of ionising photons from LAEs. The empirical calibra-
tions presented here can be easily tested with future observations
with JWST which can obtain Hα and Hβ measurements for high-
redshift LAEs.
Acknowledgements. JM acknowledges the support of a Huygens PhD fellowship
from Leiden University. We have benefited greatly from the publicly available
programming language Python, including the NumPy & SciPy (Van Der Walt
et al. 2011; Jones et al. 2001), Matplotlib (Hunter 2007) and Astropy (Astropy
Collaboration et al. 2013) packages, and the Topcat analysis program (Taylor
2013). The results and samples of LAEs used for this paper are publicly available
(see e.g. Sobral et al. 2017, 2018b) and we also provide the toy model used as a
python script.
References
An, F. X., Zheng, X. Z., Hao, C.-N., Huang, J.-S., & Xia, X.-Y. 2017, ApJ, 835,
116
Astropy Collaboration et al. 2013, A&A, 558, A33
Atek, H., Kunth, D., Hayes, M., Östlin, G., & Mas-Hesse, J. M. 2008, A&A,
488, 491
Bacon, R., Brinchmann, J., Richard, J., et al. 2015, AAP, 575, A75
Cai, Z., Fan, X., Jiang, L., et al. 2015, ApJ, 799, L19
Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682
Cardamone, C. et al. 2009, MNRAS, 399, 1191
Article number, page 7 of 9
 A&A proofs: manuscript no. Lya_SFR_SM

Cassata, P., Tasca, L. A. M., Le Fèvre, O., et al. 2015, A&A, 573, A24 Table A.1. The parameters varied in our simple toy model of 20,000
Cassata, P. et al. 2011, A&A, 525, A143 sources to interpret the observational results (see Appendix A).
Charlot, S. & Fall, S. M. 1993, ApJ, 415, 580
Chisholm, J., Gazagnes, S., Schaerer, D., et al. 2018, ArXiv e-prints
[arXiv:1803.03655] Property Minimum Maximum ∆ param.
Ciardullo, R., Zeimann, G. R., Gronwall, C., et al. 2014, ApJ, 796, 64
Dijkstra, M. 2017, ArXiv e-prints [arXiv:1704.03416]
SFR (M yr−1 ) 0.1 100 0.01 dex
Dijkstra, M. & Westra, E. 2010, MNRAS, 401, 2343 log10 (ξion /Hz erg−1 ) 24.7 25.7 0.01 dex
Ding, J., Cai, Z., Fan, X., et al. 2017, ApJ, 838, L22 fesc,LyC 0.0 0.15 0.01
Domínguez Sánchez, H. et al. 2012, MNRAS, 426, 330
Drake, A. B. et al. 2017, MNRAS, 471, 267 Lyα/Hα 8.0 9.0 0.01
Dressler, A., Henry, A., Martin, C. L., et al. 2015, ApJ, 806, 19 E(B − V) 0.0 0.5 0.01
Fynbo, J. P. U., Ledoux, C., Møller, P., Thomsen, B., & Burud, I. 2003, A&A,
407, 147 Extra fesc,Lyα 0.0 1.3 0.01
Gawiser, E., Francke, H., Lai, K., et al. 2007, ApJ, 671, 278
Gronke, M., Dijkstra, M., Trenti, M., & Wyithe, S. 2015, MNRAS, 449, 1284
Hagen, A., Zeimann, G. R., et al. 2016, ApJ, 817, 79
Harikane, Y., Ouchi, M., Shibuya, T., et al. 2017, ArXiv e-prints Appendix A: Toy-model for fesc,Lyα dependencies
[arXiv:1711.03735]
Hashimoto, T. et al. 2017, A&A, 608, A10 We construct a simple analytical toy-model to produce observ-
Hayes, M., Östlin, G., Schaerer, D., et al. 2013, ApJL, 765, L27
Hayes, M., Schaerer, D., Östlin, G., et al. 2011, ApJ, 730, 8 able Hα, UV and Lyα luminosities and EW0 from a range of in-
Hayes, M. et al. 2010, Nature, 464, 562 put physical conditions (see Table A.1). We independently sam-
Hayes, M. et al. 2014, ApJ, 782, 6
Henry, A., Scarlata, C., Martin, C. L., & Erb, D. 2015, ApJ, 809, 19 ple in steps of 0.01 or 0.01 dex combinations of SFR, fesc,LyC ,
Hu, E. M., Cowie, L. L., Barger, A. J., et al. 2010, ApJ, 725, 394
Hunter, J. D. 2007, Computing In Science & Engineering, 9, 90 case B Lyα/Hα intrinsic ratio, log10 (ξion /Hz erg−1 ), E(B − V)
Izotov, Y. I., Orlitová, I., Schaerer, D., et al. 2016a, Nature, 529, 178 with a Calzetti et al. (2000) dust law and a parameter to con-
Izotov, Y. I., Schaerer, D., Thuan, T. X., et al. 2016b, MNRAS, 461, 3683
Izotov, Y. I., Schaerer, D., Worseck, G., et al. 2018, MNRAS, 474, 4514 trol fesc,Lyα (from e.g. scattering leading to higher dust absorp-
Jones, E., Oliphant, T., Peterson, P., et al. 2001, SciPy: Open source scientific tion or scattering Lyα photons away from or into the observers’
tools for Python
Kennicutt, Jr., R. C. 1998, ARAA, 36, 189 line of sight) which acts as a further factor affecting fesc,Lyα ; see
Laursen, P., Sommer-Larsen, J., & Razoumov, A. O. 2011, ApJ, 728, 52 Table A.1 for the range in parameters explored independently.
Mainali, R., Kollmeier, J. A., Stark, D. P., et al. 2017, ApJ, 836, L14
Malhotra, S. & Rhoads, J. E. 2004, ApJL, 617, L5 We follow Kennicutt (1998) and all definitions and assumptions
Martin, C. L. & Sawicki, M. 2004, ApJ, 603, 414 mentioned in this paper. We publicly release our simple python
Matthee, J., Sobral, D., Best, P., et al. 2017a, MNRAS, 465, 3637
Matthee, J., Sobral, D., Boone, F., et al. 2017b, ApJ, 851, 145 script which can be used for similar studies and/or to study dif-
Matthee, J., Sobral, D., Darvish, B., et al. 2017c, MNRAS, 472, 772 ferent ranges in the parameter space, or conduct studies in which
Matthee, J., Sobral, D., Oteo, I., et al. 2016, MNRAS, 458, 449
Matthee, J., Sobral, D., Santos, S., et al. 2015, MNRAS, 451, 400 properties are intrinsically related/linked as one expects for real-
Meurer, G. R., Heckman, T. M., & Calzetti, D. 1999, ApJ, 521, 64 istic galaxies.
Miley, G. & De Breuck, C. 2008, AAPR, 15, 67
Momose, R., Ouchi, M., Nakajima, K., et al. 2014, MNRAS, 442, 110
Nakajima, K., Fletcher, T., Ellis, R. S., Robertson, B. E., & Iwata, I. 2018, ArXiv
e-prints [arXiv:1801.03085] Appendix A.1: The fesc,Lyα -EW0 results from a tight
Neufeld, D. A. 1991, ApJ, 370, L85
Oesch, P. A., van Dokkum, P. G., Illingworth, G. D., et al. 2015, ApJ, 804, L30 ξion -E(B − V) sequence for LAEs
Ono, Y. et al. 2012, ApJ, 744, 83
Oteo, I., Sobral, D., Ivison, R. J., et al. 2015, MNRAS, 452, 2018 We use our simple analytical model to further interpret the ob-
Ouchi, M. et al. 2008, ApJs, 176, 301 served relation between fesc,Lyα -EW0 and its tightness. We take
Ouchi, M. et al. 2009, ApJ, 696, 1164
Oyarzún, G. A., Blanc, G. A., González, V., Mateo, M., & Bailey, III, J. I. 2017, all artificially generated sources and select those that satisfy the
ApJ, 843, 133 observed relation given in Equation 3, including its scatter (see
Partridge, R. B. & Peebles, P. J. E. 1967, ApJ, 147, 868
Pritchet, C. J. 1994, PASP, 106, 1052 Figure A.1). We further restrict the sample to sources with Lyα
Rauch, M. et al. 2008, ApJ, 681, 856
Rhoads, J. E., Malhotra, S., Dey, A., et al. 2000, ApJL, 545, L85 EW0 > 25 Å. We find that along the observed fesc,Lyα -EW0 rela-
Schaerer, D. 2003, AAP, 397, 527 tion, LAEs become less affected by dust extinction as a function
Schmidt, K. B. et al. 2017, ApJ, 839, 17
Shibuya, T. et al. 2018, PASJ, 70, S15 of increasing EW0 , while ξion increases, as already shown in §3.3
Shivaei, I., Reddy, N. A., et al. 2017, ArXiv e-prints [arXiv:1711.00013] and Figure 2.
Smidt, J., Wiggins, B. K., & Johnson, J. L. 2016, ArXiv e-prints
[arXiv:1603.00888]
In the right panel of Figure A.1 we show the full parameter
Smit, R., Bouwens, R. J., Franx, M., et al. 2012, ApJ, 756, 14 range explored in ξion -E(B − V). By constraining the simulated
Sobral, D., Best, P. N., Matsuda, Y., et al. 2012, MNRAS, 420, 1926 sources with the observed fesc,Lyα -EW0 relation, we obtain a tight
Sobral, D., Best, P. N., Smail, I., et al. 2014, MNRAS, 437, 3516
Sobral, D., Matthee, J., Best, P., et al. 2017, MNRAS, 466, 1242 (±0.1 dex), linear relation between log10 ξion and E(B − V) given
Sobral, D., Matthee, J., Darvish, B., et al. 2015, ApJ, 808, 139
Sobral, D., Matthee, J., Darvish, B., et al. 2018a, ArXiv e-prints by log10 (ξion /Hz erg−1 ) ≈ −1.76 × E(B − V) + 25.6. This means
[arXiv:1802.10102] that in order for simulated sources to reproduce observations,
Sobral, D., Santos, S., Matthee, J., et al. 2018b, ArXiv e-prints LAEs should follow a very well defined ξion -E(B − V) sequence
[arXiv:1712.04451]
Song, M., Finkelstein, S. L., et al. 2014, ApJ, 791, 3 with high ξion values corresponding to very low E(B−V) (mostly
Stark, D. P., Ellis, R. S., Charlot, S., et al. 2017, MNRAS, 464, 469 at high EW0 and high fesc,Lyα ) and higher E(B − V) to lower ξion
Stark, D. P., Richard, J., Charlot, S., et al. 2015a, MNRAS, 450, 1846
Stark, D. P., Walth, G., Charlot, S., et al. 2015b, MNRAS, 454, 1393 (mostly at low EW0 and high fesc,Lyα ). Our results thus hint for
Stark, D. P. et al. 2015c, MNRAS, 450, 1846 the fesc,Lyα -EW0 to be driven by the physics (and diversity) of
Steidel, C. C., Bogosavljević, M., Shapley, A. E., et al. 2011, ApJ, 736, 160
Taylor, M. 2013, Starlink User Note, 253 young and metal poor stellar populations and their evolution.
Tilvi, V., Pirzkal, N., Malhotra, S., et al. 2016, ApJ, 827, L14
Trainor, R. F., Steidel, C. C., Strom, A. L., & Rudie, G. C. 2015, ApJ, 809, 89
Trainor, R. F., Strom, A. L., Steidel, C. C., & Rudie, G. C. 2016, ApJ, 832, 171
van Breukelen, C., Jarvis, M. J., & Venemans, B. P. 2005, MNRAS, 359, 895 Appendix B: Data used for the high-redshift
Van Der Walt, S., Colbert, S. C., & Varoquaux, G. 2011, Computing in Science
& Engineering, 13, 22 comparison between UV and Lyα SFRs
Vanzella, E., Pentericci, L., Fontana, A., et al. 2011, ApJ, 730, L35
Verhamme, A., Orlitová, I., Schaerer, D., et al. 2017, A&A, 597, A13 Table A.2 provides the data used for Figure 4, including indi-
Verhamme, A., Schaerer, D., Atek, H., & Tapken, C. 2008, A&A, 491, 89 vidual measurements per source, their name and reference. Note
Wisotzki, L., Bacon, R., et al. 2016, A&A, 587, A98
Yang, H., Malhotra, S., Gronke, M., et al. 2016, ApJ, 820, 130 that the data is taken from a compilation from Matthee et al.
Yang, H., Malhotra, S., Gronke, M., et al. 2017, ApJ, 844, 171 (2017c) with minor modifications for a few LAEs, as a indicated
Zabl, J., Nørgaard-Nielsen, H. U., Fynbo, J. P. U., et al. 2015, MNRAS, 451,
2050 in Table A.2.

Article number, page 8 of 9

 Sobral & Matthee: Predicting Lyα escape fractions with a simple observable

Fig. A.1. Left: The predicted relation between fesc,Lyα and Lyα EW0 for our toy model, which shows little to no correlation by sampling all
physical parameters independently (see Table A.1). We also show the observed range (≈ ±3σ) which is well constrained at z ∼ 0 − 2.6. We use
simulated sources that are consistent with observations of LAEs to explore the potential reason behind the observed tight fesc,Lyα -EW0 correlation
for LAEs. Right: By restricting our toy model to the observed relation and its scatter, we find a relatively tight ξion -E(B − V) sequence for LAEs
(EW0 > 20 − 25 Å): log10 (ξion /Hz erg−1 ) ≈ −1.76 × E(B − V) + 25.6. The highest observed EW0 correspond to the highest ξion and the lowest
E(B − V), while lower EW0 leads to a lower ξion and a higher E(B − V). Our results thus show that the tight fesc,Lyα -EW0 correlation for LAEs at
z ∼ 0 − 2.6 is likely driven by a ξion -E(B − V) sequence that may be related with important physics such as the age of the stellar populations, their
metallicity, dust production and how those evolve together.

Table A.2. Application to high redshift UV-continuum and Lyα selected LAEs (see compilation by Matthee et al. 2017c). Errors on Lyα luminosity
and EW0 are assumed to be ≈ 0.1 dex, while errors on MUV are taken as ≈ 0.2 dex. We compute the UV SFRs (SFRUV , dust corrected) using
Kennicutt (1998) and β = −1.6 ± 0.2 for UV-selected and β = −1.9 ± 0.2 for Lyα selected sources. Lyα SFRs (SFRLyα ; calibrated to be dust-
corrected) are computed with our Equation 4. Notes: 1: EW0 have been recomputed and rest-framed when compared to original reference. 2: MUV
have been recomputed when compared to original reference. 3: Values used are from Zabl et al. (2015). 4: Computed as in Matthee et al. (2017b).
This table is also provided in fits format.

Name z log10 (LLyα ) EW0 MUV SFRUV SFRLyα Reference

(UV selected) [erg s−1 ] [Å] [mag] [M yr−1 ] [M yr−1 ]
A383-5.2 6.03 42.8 138 −19.3 10+2
−3 11+2
−3 Stark et al. (2015c)
RXCJ2248.7-4431-ID3 6.11 42.5 40 −20.1 21+5
−6 16+4
−5 Mainali et al. (2017)
RXCJ2248.7-4431 6.11 42.9 68 −20.2 23+5
−7 25+5
−7 Schmidt et al. (2017)
SDF-46975 6.84 43.2 43 −21.5 76+18
−23 76+17
−24 Ono et al. (2012)
IOK-1 6.96 43.0 42 −21.3 63+14
−19 57+13
−18 Ono et al. (2012)
BDF-521 7.01 43.0 64 −20.6 34+8
−10 34+7
−9 Cai et al. (2015)
A1703 zd6 7.04 42.5 65 −19.3 10+2
−3 10+2
−3 Stark et al. (2015b)
BDF-3299 7.11 42.8 50 −20.6 33+8
−10 30+6
−9 Vanzella et al. (2011)
GLASS-stack 7.20 43.0 210 −19.7 15+3
−4 10+2
−3 Smidt et al. (2016)
EGS-zs8-2 7.48 42.7 9 −21.9 110+25
−32 103+45
−204 Stark et al. (2015a)
FIGS GN1 1292 7.51 42.8 49 −21.2 58+13
−17 31+7
−9 Tilvi et al. (2016)
GN-108036 7.21 43.2 33 −21.8 101+24
−30 99+24
−37 Stark et al. (2015a)
EGS-zs8-1 7.73 43.1 21 −22.1 131+30
−40 124+36
−71 Oesch et al. (2015)
(Lyα selected)
SR61 5.68 43.4 210 −21.1 30+7
−9 26+5
−6 Matthee et al. (2017c)
Ding-3 5.69 42.8 62 −20.9 25+6
−8 25+5
−7 Ding et al. (2017)
Ding-4 5.69 42.3 106 −20.5 18+4
−5 4+1
−1 Ding et al. (2017)
Ding-5 5.69 43.2 79 −21.0 28+6
−8 44+9
−12 Ouchi et al. (2008)
Ding-1 5.70 43.0 21 −22.2 85+20
−25 104+30
−59 Ding et al. (2017)
J2334542 5.73 43.7 210 −21.5 44+10
−13 51+10
−12 Shibuya et al. (2018)
J021835 5.76 43.7 107 −21.7 53+12
−16 93+19
−24 Shibuya et al. (2018)
VR71 6.53 43.4 35 −22.5 111+26
−34 149+35
−53 Matthee et al. (2017c)
J1621262 6.54 43.9 99 −22.8 146+34
−44 170+34
−44 Shibuya et al. (2018)
J160234 6.58 43.5 81 −21.9 64+15
−19 88+18
−23 Shibuya et al. (2018)
Himiko3 6.59 43.6 65 −22.1 77+18
−23 143+30
−39 Ouchi et al. (2009)
CR74 6.60 43.9 211 −22.2 84+19
−26 87+17
−22 Sobral et al. (2015)

Article number, page 9 of 9